Artificial Intelligence-Enabled, Fully Automated Detection of Cardiac Amyloidosis Using Electrocardiograms and Echocardiograms

Abstract

<jats:p>Although individually uncommon, rare diseases collectively affect over 350 million patients worldwide and are increasingly the target of therapeutic development efforts. Unfortunately, the pursuit and use of such therapies have been hindered by a common challenge: patients with specific rare diseases are difficult to identify, especially if the conditions resemble more prevalent disorders. Cardiac amyloidosis is one such rare disease, which is characterized by deposition of misfolded proteins within the heart muscle resulting in heart failure and death. In recent years, specific therapies have emerged for cardiac amyloidosis and several more are under investigation, but because cardiac amyloidosis is mistaken for common forms of heart failure, it is typically diagnosed late in its course. As a possible solution, artificial intelligence methods could enable automated detection of rare diseases, but model performance must address low disease prevalence. Here we present an automated multi-modality pipeline for cardiac amyloidosis detection using two neural-network models; one using electrocardiograms (ECG) and the second using echocardiographic videos as input. These models were trained and validated on 3 and 5 academic medical centers (AMC), respectively, in the United States and Japan. Both models had excellent accuracy for detecting cardiac amyloidosis with C-statistics of 0.85-0.92 and 0.91-1.00 for the ECG and echocardiography models, respectively, with the latter outperforming expert diagnosis. Simulating deployment on 13,906 and 7775 patients with ECG-echocardiography paired data for AMC2 and AMC3 indicated a positive predictive value (PPV) for the ECG model of 4% and 3% at 61% and 54% recall, respectively. Pre-screening with ECG enhanced the echocardiography model performance from PPV 23% and 20% to PPV 58% and 53% at 64% recall, respectively for AMC2 and AMC3. In conclusion, we have developed a robust pipeline to augment detection of cardiac amyloidosis, which should serve as a generalizable strategy for other rare and intermediate frequency cardiac diseases with established or emerging therapies.</jats:p>

Artificial intelligence in echocardiography: detection, functional evaluation, and disease diagnosis

“…In the latest research, a human-interpretation-free machine learning pipeline based on the combination of ECG and echocardiography had been developed to detect cardiac amyloidosis. Multicenter study had confirmed that the artificial intelligenceenabled fully automated detection model outperformed interpretation by expert cardiologists in the diagnosis of cardiac amyloidosis [38]. On the other hand, speckle tracking imaging technology is widely used for the diagnosis of cardiomyopathy.…”

The Future Role of High-Performance Computing in Cardiovascular Medicine and Science -Impact of Multi-Dimensional Data Analysis-

“…Thus, AI has the potential to evolve "expert intuition" into quantifiable science with the potential to improve the quality and accuracy of personalized medicine and individual-based clinical decision making. A few such examples converting "intuition" into more objective, quantitative approaches include: AI-enhanced electrocardiography (ECG) [4][5][6][7] ; ultrasonic echocardiography (UCG) 8,9) ; and serial bio-marker analyses 10) ; each of which provide patient-level quantitative CV disease risk prediction using multi-dimensional input data.…”

Prospects for cardiovascular medicine using artificial intelligence

“…In the group using ECG AI, the detection rate of EF reduction increased by ∼30%. Goto et al developed an ECG AI for diagnosing cardiac amyloidosis [22] . In the model created using ECG data of 3191 cases, the performance was good, with the AUC of 0.91.…”